Results

So here are the results of the stereo matching. The original image(left view) is shown on the left. The results of our processing is shown on the right. Darker pixels indicate areas that are closer and lighter pixels indicate images that are farther away. We see that the algorithm pretty much works!

First, here is a simple image. A pair of views of points(crosses). The top "+" is at the same position in both views and the disparity increases as you go down. This means that the "+" located at the top should be farthest away.

And here is our image of the spheres

A couple of observations:

We do not get much correlation towards the centers of the objects. This may be beacuse they are slightly shinier and match the color of the floor.

You can also see some random sploches which correspond to the texture of the floor. To get rid of these a more sophisticated peak matching algorithm is needed or maybe multi-scale peak matching(where you wouldn't see the texture of the floor far away).